If evidence can be generated, what happens to trust?

A photo used to settle things. Now it might just start them.

A familiar problem, but not the same one

Last week, the BBC reported a rise in fraud claims linked to manipulated images. At a glance, it sounds like a familiar story. AI tools make it easier to edit photos, which makes it easier to fake evidence. Insurers respond. Systems adapt. The cycle continues.

But there’s a shift underneath that feels easy to miss. It’s not just that evidence can be faked more easily. It’s that it can now be created quickly enough, and convincingly enough, to influence decisions in real time.

When false evidence gets acted on

A recent example makes that clearer. Someone generated an image of an escaped zoo wolf. Authorities believed it was real, and the search operation changed direction based on that image. Time, resources, and attention were redirected toward something that didn’t exist.

This wasn’t a coordinated attack. It wasn’t even particularly sophisticated. It was just someone creating something that looked plausible. And that was enough.

What’s changed

Before, fake evidence tended to have limits. It took time to create. It was often easier to question. And even when it spread, it didn’t always lead directly to action.

Now, the conditions are different. Evidence can be generated quickly, cheaply, and in context. It doesn’t need to convince everyone. It just needs to reach the right person, at the right moment, with enough credibility to be acted on.

From misinformation to decision-making

That changes the nature of the risk. Because the impact isn’t just about misinformation spreading. It’s about decisions being made.

In the insurance example, that might mean false claims being processed or investigated. In the wolf case, it meant real-world resources being redirected. In other situations, it could influence security responses, operational decisions, or public communication.

In one discussion around manipulated insurance images, people were asked to tell the difference between real and AI-generated photos. The results weren’t reassuring. As Nicola put it, “I scored a measly 3 or 4 out of 10 but was about average with that.”

Even when people knew what they were looking for, it wasn’t easy to tell.

The shift isn’t just about whether something is true. It’s about when that question gets asked. In many cases, the decision comes first. The image is seen, the response is triggered, and only later does the question of accuracy surface. By then, the impact has already happened.

Different domains, same underlying shift. AI doesn’t just generate content. It exposes the assumptions our systems rely on. In this case, a simple one: that what we can see is a reliable signal of what happened.

What replaces evidence?

If photos, audio, or documents can no longer be taken at face value, what replaces them as evidence?

It also changes something more subtle. Trust becomes less of an assumption, and more of something that has to be actively built. As Natalie put it, “Trust is being eroded across society – and in an AI world, that changes everything.”

And more importantly, what happens when decisions are made before doubt has a chance to catch up?